The risk gradient

Ann Gibbons reports Gibbons:diabetes:2011 from the International Congress of Human Genetics, on papers that examine GWAS risk alleles for type 2 diabetes: “Diabetes Genes Decline Out of Africa” (paywall).

At the poster session, Stanford graduate student Erik Corona stood in front of a Google Earth map of the world that he finds surprising. On this map he had plotted the frequency of 12 gene variants known to be associated with type 2 diabetes in 51 populations from Australia to Zaire. It shows a clear gradient of red to green from west to east, from Africa to Asia, Corona says (see map). Something strange is going on with type 2 diabetes.

This is of course a challenging problem because risk alleles identified in one population may not replicate in other populations. The most well-known example is ApoE4, strongly associated with Alzheimer’s Disease in Europeans, but not in Africans. More generally, looking at a set of risk variants that are identified in one population introduces an ascertainment bias that constrains their likely frequencies in other populations. An allele is more likely to yield a statistically significant association with a trait if the allele is not too rare. If we take many alleles associated with a trait, we’re likely to see some gradient across populations due to this bias alone.

Hidden ascertainment bias is a problem we run up against quite a lot. It may not apply in this case, depending on where the risk alleles were identified, in particular since many risk alleles for type 2 diabetes appear to be linked to recent positive selection (explaining why I got interested).